PHASE: Personalized Head-based Automatic Simulation for Electromagnetic Properties in 7T MRI

The official implementation of PHASE.

First Figure

Abstract

Accurate and individualized human head models are becoming increasingly important for electromagnetic (EM) simulations. These simulations depend on precise anatomical representations to realistically model electric and magnetic field distributions, particularly when evaluating Specific Absorption Rate (SAR) within safety guidelines. We introduce Personalized Head-based Automatic Simulation for EM properties (PHASE), an automated open-source toolbox that generates high-resolution, patient-specific head models for EM simulations using paired T1-weighted (T1w) magnetic resonance imaging (MRI) and computed tomography (CT) scans with 13 tissue labels. To evaluate the performance of PHASE models, we conduct semi-automated segmentation and EM simulations on 15 real human patients, serving as the gold standard reference. The PHASE model achieved comparable global SAR and localized SAR averaged over 10 grams of tissue (SAR-10g), demonstrating its potential as a promising tool for generating large-scale human model datasets in the future.

Hightlights

We propose PHASE, an automatic toolbox which can automatically generate personalized human head models for EM simulation usage from pairs of real MRI and CT scans.
We perform a comparative analysis against the outputs of manually refined reference models and existing models, evaluating the SAR accuracy through EM simulations.
We explore tissue groupings to evaluate how varying levels of anatomical details affect the EM simulation results of ultrahigh field RF transmit coil.
We release PHASE as an open resource toolbox to support research in medical physics, including applications in RF coil design and EM safety evaluation.

Quick start

Brain segmentation and mapping

SLANT brain segmentation [1] is applied to segment detailed brain region. Refer to SLANT to run the trained model on your T1w MRI volume and get the segmented .nii.gz final results.

A brain mapping from 133 anatomical labels to 8 tissue groups with distinct electrical properties is applied to the segmented brain. Run:

python SLANT_label_mapping.py --input_dir your/input/dict --output_dir your/output/dict --mapping_file braincolor_hierarchy_STAPLE.txt

Other tissues

SimNIBS [2] and GRACE [3] are used to segment and fill in the other parts of the human head.

Automatic correction

With segmented brain from SLANT, other tissues from SimNIBS and GRACE, bones from registered CT, run to combine and perform correction:

python combination.py

Model construction: from `.nii` to `.raw`

A script transforming .nii file to .raw file which is importable for most simulation software are provided.

python nii_to_raw.py

An example of a header .txt file needed for .raw when importing is provided. The grid extent and spatial steps need to be refined to your models.

Reference

[1] Huo, Yuankai, et al. “3D whole brain segmentation using spatially localized atlas network tiles.” NeuroImage 194 (2019): 105-119.
[2] Puonti, Oula, et al. “Accurate and robust whole-head segmentation from magnetic resonance images for individualized head modeling.” Neuroimage 219 (2020): 117044.
[3] Stolte, Skylar E., et al. “Precise and rapid whole-head segmentation from magnetic resonance images of older adults using deep learning.” Imaging Neuroscience 2 (2024): 1-21.

Contact

For any questions or discussion, email us.